1.4. Lecture 4

1.4.1. Schur’s Lemma

We move on to more theoretical considerations. Recall that if GG is a group and (π,V)(\pi,V) is a representation of GG, then HomG(V)=HomG(V,V)\operatorname{Hom}_{G}(V)=\operatorname{Hom}_{G}(V,V) is the vector space of linear maps from VV to VV that commute with every π(g)\pi(g) (for all gGg\in G).

Lemma 1.4.1 (Schur’s Lemma, 1907).

If (π,V)(\pi,V) is an irreducible finite-dimensional complex representation of a group GG, then

HomG(V)=Id.\operatorname{Hom}_{G}(V)={\mathbb{C}}{\,\mathrm{Id}}.
Proof.

Let T(0)HomG(V)T(\neq 0)\in\operatorname{Hom}_{G}(V). Since VV is a finite-dimensional vector space over {\mathbb{C}}, TT has an eigenvalue λ\lambda\in{\mathbb{C}}. Consider the morphism TλIdT-\lambda\,{\,\mathrm{Id}}. Then for any gGg\in G,

π(g)(TλId)=π(g)Tλπ(g)=Tπ(g)λπ(g)=(TλId)π(g);\pi(g)(T-\lambda{\,\mathrm{Id}})=\pi(g)T-\lambda\pi(g)=T\pi(g)-\lambda\pi(g)=(% T-\lambda{\,\mathrm{Id}})\pi(g);

TλIdT-\lambda{\,\mathrm{Id}} is therefore in HomG(V)\operatorname{Hom}_{G}(V). By Lemma 1.2.9, (π,ker(TλId))\big{(}\pi,\ker(T-\lambda{\,\mathrm{Id}})\big{)} is a subrepresentation of (π,V)(\pi,V). The subspace ker(TλId)V\ker(T-\lambda{\,\mathrm{Id}})\subseteq V is non-zero, since any λ\lambda-eigenvector of TT is in ker(TλId)\ker(T-\lambda{\,\mathrm{Id}}), and so (π,V)(\pi,V) being irreducible then gives V=ker(TλId)V=\ker(T-\lambda{\,\mathrm{Id}}), hence T=λIdIdT=\lambda{\,\mathrm{Id}}\in{\mathbb{C}}{\,\mathrm{Id}}. ∎

Corollary 1.4.2.

Let (π,V)(\pi,V) and (ρ,W)(\rho,W) be two irreducible finite-dimensional complex representations of a group GG. Then

dimHomG(V,W)={1 if (π,V)(ρ,W)0 otherwise.\operatorname{dim}\operatorname{Hom}_{G}(V,W)=\begin{cases}1&\text{ if $(\pi,V% )\cong(\rho,W)$}\\ 0&\text{ otherwise}.\end{cases}
Proof.

We start by assuming that the representations are not isomorphic. Given THomG(V,W)T\in\operatorname{Hom}_{G}(V,W), by Lemma 1.2.9, (π,ker(T))(\pi,\ker(T)) is a subrepresentation of (π,V)(\pi,V) and (ρ,im(T))(\rho,{\mathrm{im}}(T)) is a subrepresentation of (ρ,W)(\rho,W). By assumption, (π,V)(\pi,V) and (ρ,W)(\rho,W) are irreducible, hence

ker(T)=0orVandim(T)=0orW.\ker(T)=0\;\mathrm{or}\;V\quad\mathrm{and}\quad{\mathrm{im}}(T)=0\;\mathrm{or}% \;W.

If im(T)=0{\mathrm{im}}(T)=0, then T=0T=0, and we are done. If im(T)=W{\mathrm{im}}(T)=W, then ker(T)V\ker(T)\neq V, hence ker(T)=0\ker(T)=0. The map TT is therefore both injective (ker(T)=0\ker(T)=0) and surjective (im(T)=W{\mathrm{im}}(T)=W), i.e. invertible. This contradicts the assumption that the representations are non-isomorphic; hence we must have im(T)=0{\mathrm{im}}(T)=0.

Assume now that the representations are isomorphic. Then by assumption there exists an invertible element THomG(V,W)T\in\operatorname{Hom}_{G}(V,W). Let SS be any other element of HomG(V,W)\operatorname{Hom}_{G}(V,W). We then consider the map T1S:VVT^{-1}S:V\rightarrow V. By Problem 1.2.3, T1HomG(W,V)T^{-1}\in\operatorname{Hom}_{G}(W,V), hence for any gGg\in G, we have

T1Sπ(g)=T1ρ(g)S=π(g)T1S,T^{-1}S\pi(g)=T^{-1}\rho(g)S=\pi(g)T^{-1}S,

hence T1SHomG(V)T^{-1}S\in\operatorname{Hom}_{G}(V). By Lemma 1.4.1, HomG(V)=Id\operatorname{Hom}_{G}(V)={\mathbb{C}}{\,\mathrm{Id}}, hence T1S=λIdT^{-1}S=\lambda{\,\mathrm{Id}} for some λ\lambda\in{\mathbb{C}}. It follows that STS\in{\mathbb{C}}T, showing that dimHomG(V,W)=1\operatorname{dim}\operatorname{Hom}_{G}(V,W)=1. ∎

1.4.2. Abelian groups

Schur’s lemma has a particularly striking application to Abelian groups:

Theorem 1.4.3.

Let GG be an Abelian group. Then every finite-dimensional irreducible complex representation of GG is 1-dimensional.

Proof.

Let (ρ,V)(\rho,V) be an irreducible representation of GG. For hGh\in G, set Th=ρ(h)GL(V)T_{h}=\rho(h)\in\operatorname{GL}(V). Then ThHomG(V)T_{h}\in\operatorname{Hom}_{G}(V), since it commutes with ρ(g)\rho(g) for all gGg\in G. Indeed,

Thρ(g)=ρ(h)ρ(g)=ρ(hg)=ρ(gh)=ρ(g)ρ(h)=ρ(g)Th.T_{h}\rho(g)=\rho(h)\rho(g)=\rho(hg)=\rho(gh)=\rho(g)\rho(h)=\rho(g)T_{h}.

Hence by Schur’s Lemma, Th=ρ(h)T_{h}=\rho(h) acts by a scalar χ(h)\chi(h) on VV:

ρ(h)𝐯=χ(h)𝐯\rho(h){\bf v}=\chi(h){\bf v}

for all 𝐯V{\bf v}\in V and a non-zero scalar χ(h)\chi(h). But now, any non-zero 𝐯V{\bf v}\in V spans a GG-invariant subspace. Since VV is irreducible, this implies that VV is 1-dimensional, and χ=ρ\chi=\rho is a homomorphism G×G\rightarrow\mathbb{C}^{\times}. ∎

Remark 1.4.4.

It is possible to give an alternative proof of this using the fact from linear algebra that any commuting set of linear maps from a finite-dimensional vector space to itself has a simultaneous eigenvector.

A homomorphism χ:G×\chi:G\rightarrow\mathbb{C}^{\times} is often called a character (though this will later cause an unfortunate clash of notation). If GG is Abelian, then the group

G^={homomorphisms χ:G×}\widehat{G}=\{\text{homomorphisms }\chi:G\rightarrow\mathbb{C}^{\times}\}

is called the character group, or dual group, of GG. It is a group under the operation (χ1χ2)(g)=χ1(g)χ2(g)(\chi_{1}\chi_{2})(g)=\chi_{1}(g)\chi_{2}(g).

Example 1.4.5.

Let G=CnG=C_{n} be a cyclic group of order nn. Then G^Cn\widehat{G}\cong C_{n}. Indeed, pick a generator gg of GG and let ω=e2πi/n\omega=e^{2\pi i/n} be a primitive nn-th root of unity. Then a character χ\chi of GG is determined uniquely by χ(g)\chi(g), which must be an nn-th root of unity ωa\omega^{a}. If we let χaG^\chi_{a}\in\hat{G} be the homomorphism such that χa(g)=ωa\chi_{a}(g)=\omega^{a}, then the map

aχaa\mapsto\chi_{a}

determines a group isomorphism /nG^{\mathbb{Z}}/n{\mathbb{Z}}\rightarrow\widehat{G} . This is a homomorphism because

χa+b(g)=ωa+b=ωaωb=χa(g)χb(g).\chi_{a+b}(g)=\omega^{a+b}=\omega^{a}\omega^{b}=\chi_{a}(g)\chi_{b}(g).

In fact, if GG is any finite Abelian group, then G^G\widehat{G}\cong G. You can prove this using the cyclic case and the fundamental theorem of finite Abelian groups.

For arbitrary groups GG the same method of proof gives:

Proposition 1.4.6.

Let (ρ,V)(\rho,V) be an irreducible finite-dimensional representation of GG and let

Z=Z(G)={zG|zg=gz for all gG}Z=Z(G)=\{z\in G\,|\,zg=gz\text{ for all $g\in G$}\}

be the center of GG. Then ZZ acts on VV as a character: there is a character χ:Z×\chi:Z\rightarrow{\mathbb{C}}^{\times} such that

ρ(z)𝐯=χ(z)𝐯.\rho(z){\bf v}=\chi(z){\bf v}.

for all zZz\in Z and 𝐯V{\bf v}\in V.

We call χ\chi the central character of ρ\rho.

Proof.

See Problem 1.4.3 below. ∎

Finally, we can use our classification of the irreducible representations of Abelian groups to get a bound on the dimension of the irreducible representations of any finite group.

Proposition 1.4.7.

Let GG be a finite group, AA be an Abelian subgroup of GG, and (π,V)(\pi,V) be an irreducible representation of GG. Then

dimV|G||A|=[G:A].\operatorname{dim}V\leq\frac{|G|}{|A|}=[G:A].
Proof.

Restricting the representation to AA, we find an irreducible AA-subrepresentation WW of VV (cf. Problem 1.4.3). By Theorem 1.4.3, WW is 1-dimensional, spanned by a vector 𝐯{\bf v}. There is then a character χ\chi of AA such that

π(h)𝐯=χ(h)𝐯\pi(h){\bf v}=\chi(h){\bf v}

for all hAh\in A. Now, {π(g)𝐯|gG}\{\pi(g){\bf v}\,|\,g\in G\} spans a non-zero subrepresentation of VV, hence is equal to VV by irreducibility. Write g1A,g2A,,grAg_{1}A,g_{2}A,\ldots,g_{r}A for the left cosets of AA in GG, where r=[G:A]r=[G:A]. Then for any hAh\in A, we have

π(gih)𝐯=π(gi)π(h)𝐯=π(gi)χ(h)𝐯=χ(h)(π(gi)𝐯)\pi(g_{i}h){\bf v}=\pi(g_{i})\pi(h){\bf v}=\pi(g_{i})\chi(h){\bf v}=\chi(h)(% \pi(g_{i}){\bf v})

But this implies that V=span{π(g)v|gG}V=\mathrm{span}\{\pi(g)v\,|\,g\in G\} is already spanned by

{π(gi)v| 1ir},\{\pi(g_{i})v\,|\,1\leq i\leq r\},

so has dimension at most r=[G:A]r=[G:A]. ∎

Example 1.4.8.

The group DnD_{n} has an abelian subgroup CnC_{n} of index 2, and so every irreducible representation of DnD_{n} has dimension at most 22.

1.4.3. Exercises

.

Problem 29. Let (π,V)(\pi,V) be a finite-dimensional representation of a group GG. Show that there is an irreducible subrepresentation (π,W)(\pi,W) of (π,V)(\pi,V).

Problem 30. Let (π,V)(\pi,V) be an irreducible finite-dimensional representation of a group GG. Denoting the centre of GG by Z(G)Z(G) (i.e. Z(G)={gG|gh=hghG}Z(G)=\{g\in G\,|\,gh=hg\quad\forall h\in G\}), show that there exists a homomorphism ψ:Z(G)×\psi:Z(G)\rightarrow{\mathbb{C}}^{\times} such that

π(g)=ψ(g)Id\pi(g)=\psi(g)\,\mathrm{Id}

for all gZ(G)g\in Z(G). Hint: adapt the proof of Theorem 1.4.3.